Vectara MCP server

The Vectara MCP server provides AI agents with access to Vectara’s semantic search and retrieval-augmented generation (RAG) APIs.

When should you use this server

Use Vectara MCP when you want an agent to:

• Run semantic search over enterprise documents.
• Perform RAG queries that return both search results and a synthesized answer.
• Ground responses in organization-specific data with minimal setup.

Tools provided

ask_vectara

Run a RAG query using Vectara. Returns search results along with a generated response.

Arguments

• query (string, required) — the user query to run.
• corpus_keys (list[string], required) — corpus keys to search; the user must provide one or more.
• api_key (string, required) — the Vectara API key.
• n_sentences_before (int, optional, default=2) — sentences before a hit to include in context.
• n_sentences_after (int, optional, default=2) — sentences after a hit to include in context.
• lexical_interpolation (float, optional, default=0.005) — balance between semantic and lexical match.
• max_used_search_results (int, optional, default=10) — max number of search results to use.
• generation_preset_name (string, optional, default=vectara-summary-table-md-query-ext-jan-2025-gpt-4o) — generation preset.
• response_language (string, optional, default=eng) — language of the generated response.

Returns

• Generated answer (string)
• Supporting search results (list of documents/snippets)

Example usage

• “Summarize the refund policy from the Finance corpus.”
• “Answer FAQs about GDPR using the Compliance corpus.”

---

search_vectara

Run a semantic search query without generation. Returns only the most relevant search results.

Arguments

• query (string, required) — the user query to run.
• corpus_keys (list[string], required) — corpus keys to search.
• api_key (string, required) — the Vectara API key.
• n_sentences_before (int, optional, default=2) — sentences before a hit to include.
• n_sentences_after (int, optional, default=2) — sentences after a hit to include.
• lexical_interpolation (float, optional, default=0.005) — balance between semantic and lexical match.

Returns

• Matching search results with context snippets.

Example usage

• “Find all product docs mentioning SAML login.”
• “Search engineering notes for ‘distributed caching.’”

Source